Ensemble ANN with two ensembles a MLP and LSTM network for wind speed prediction from data collected with windGURU stations. This repository contains code that was used for conducting experiments in master thesis submitted in November 2018 on the University of Liechtenstein.
To run the predictions run directly one of theese files:
trains the LSMT network on shifted measurement data and predict on splitted test dataset and draws the plot
trains the MLP network on inputs from WRF forecast and expected outputs of measurements, predicts on splitted test dataset and draws the plot
trains the same architectures as in single-LSTM.py and single-MLP.py on a training set and predicts for a test set. Then trains the ensemble MPL network on a training set and predicts from the two produced test sets and then draws the plot and calculates NRMSE and outputs it to the console.
The folder contains the measurements and WRF9km forecasts for four experimental spots. Note: Files only cointain structures of the .csv files, permission to release the datasets have not been gained yet.
The measurements files are pretty straightforward, but the WRF forecast files cointain under the rows with id_var
id_var=1 : wind speed
id_var=2 : wind direction
id_var=3 : temperature
The forecast is placed in the h[n] collumn
column h1 : Forecast for hour +1 from the initialization time in 'datum' collumn
column h2 : Forecast for hour +2 from the initialization time in 'datum' collumn
column h3 : Forecast for hour +3 from the initialization time in 'datum' collumn
...
...
column h78 : Forecast for hour +78 from the initialization time in 'datum' collumn